…port

Currently the only mode is the existing behaviour, however this
opens the door to other approaches.

This is unlikely to be common, however it provides an escape hatch
for advanced users who want to do their own thing.

This is slightly wassteful, but should mean we cheaply get good
coverage over all the interesting model relations we come up with.

Sequences don't nicely fit into one of just schema or data, they're
somewhat inherently both. Given that Django's "loaddata" over-writes
existing rows in tables, it seems reasonable to do something similar
for sequences -- even if that means we actually drop the sequence
and fully re-create it.

This happens in self.dump_data_for_import, likely I missed this in 78f06a3.

This introduces a util to simplify this and adds usage to a couple
of places which had been missed.

Essentially we get a merging of the two sources of data.

Importing, even in the 'none' case, is potentially destructive so
we want to give the user a chance to bail.

Also includes docs for the Postgres sequences extra.